rtoss - Rev 73

Subversion Repositories:
Rev:
/*
 * QEMU posix-aio emulation
 *
 * Copyright IBM, Corp. 2008
 *
 * Authors:
 *  Anthony Liguori   <aliguori@us.ibm.com>
 *
 * This work is licensed under the terms of the GNU GPL, version 2.  See
 * the COPYING file in the top-level directory.
 *
 */


#include <sys/ioctl.h>
#include <sys/types.h>
#include <pthread.h>
#include <unistd.h>
#include <errno.h>
#include <time.h>
#include <signal.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>

#include "qemu-queue.h"
#include "osdep.h"
#include "qemu-common.h"
#include "block_int.h"

#include "block/raw-posix-aio.h"


struct qemu_paiocb {
    BlockDriverAIOCB common;
    int aio_fildes;
    union {
        struct iovec *aio_iov;
        void *aio_ioctl_buf;
    };
    int aio_niov;
    size_t aio_nbytes;
#define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
    int ev_signo;
    off_t aio_offset;

    QTAILQ_ENTRY(qemu_paiocb) node;
    int aio_type;
    ssize_t ret;
    int active;
    struct qemu_paiocb *next;
};

typedef struct PosixAioState {
    int rfd, wfd;
    struct qemu_paiocb *first_aio;
} PosixAioState;


static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;
static pthread_cond_t cond = PTHREAD_COND_INITIALIZER;
static pthread_t thread_id;
static pthread_attr_t attr;
static int max_threads = 64;
static int cur_threads = 0;
static int idle_threads = 0;
static QTAILQ_HEAD(, qemu_paiocb) request_list;

#ifdef CONFIG_PREADV
static int preadv_present = 1;
#else
static int preadv_present = 0;
#endif

static void die2(int err, const char *what)
{
    fprintf(stderr, "%s failed: %s\n", what, strerror(err));
    abort();
}

static void die(const char *what)
{
    die2(errno, what);
}

static void mutex_lock(pthread_mutex_t *mutex)
{
    int ret = pthread_mutex_lock(mutex);
    if (ret) die2(ret, "pthread_mutex_lock");
}

static void mutex_unlock(pthread_mutex_t *mutex)
{
    int ret = pthread_mutex_unlock(mutex);
    if (ret) die2(ret, "pthread_mutex_unlock");
}

static int cond_timedwait(pthread_cond_t *cond, pthread_mutex_t *mutex,
                           struct timespec *ts)
{
    int ret = pthread_cond_timedwait(cond, mutex, ts);
    if (ret && ret != ETIMEDOUT) die2(ret, "pthread_cond_timedwait");
    return ret;
}

static void cond_signal(pthread_cond_t *cond)
{
    int ret = pthread_cond_signal(cond);
    if (ret) die2(ret, "pthread_cond_signal");
}

static void thread_create(pthread_t *thread, pthread_attr_t *attr,
                          void *(*start_routine)(void*), void *arg)
{
    int ret = pthread_create(thread, attr, start_routine, arg);
    if (ret) die2(ret, "pthread_create");
}

static size_t handle_aiocb_ioctl(struct qemu_paiocb *aiocb)
{
        int ret;

        ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
        if (ret == -1)
                return -errno;

        /*
         * This looks weird, but the aio code only consideres a request
         * successfull if it has written the number full number of bytes.
         *
         * Now we overload aio_nbytes as aio_ioctl_cmd for the ioctl command,
         * so in fact we return the ioctl command here to make posix_aio_read()
         * happy..
         */

        return aiocb->aio_nbytes;
}

static size_t handle_aiocb_flush(struct qemu_paiocb *aiocb)
{
    int ret;

    ret = qemu_fdatasync(aiocb->aio_fildes);
    if (ret == -1)
        return -errno;
    return 0;
}

#ifdef CONFIG_PREADV

static ssize_t
qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
{
    return preadv(fd, iov, nr_iov, offset);
}

static ssize_t
qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
{
    return pwritev(fd, iov, nr_iov, offset);
}

#else

static ssize_t
qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
{
    return -ENOSYS;
}

static ssize_t
qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
{
    return -ENOSYS;
}

#endif

static size_t handle_aiocb_rw_vector(struct qemu_paiocb *aiocb)
{
    size_t offset = 0;
    ssize_t len;

    do {
        if (aiocb->aio_type & QEMU_AIO_WRITE)
            len = qemu_pwritev(aiocb->aio_fildes,
                               aiocb->aio_iov,
                               aiocb->aio_niov,
                               aiocb->aio_offset + offset);
         else
            len = qemu_preadv(aiocb->aio_fildes,
                              aiocb->aio_iov,
                              aiocb->aio_niov,
                              aiocb->aio_offset + offset);
    } while (len == -1 && errno == EINTR);

    if (len == -1)
        return -errno;
    return len;
}

static size_t handle_aiocb_rw_linear(struct qemu_paiocb *aiocb, char *buf)
{
    size_t offset = 0;
    size_t len;

    while (offset < aiocb->aio_nbytes) {
         if (aiocb->aio_type & QEMU_AIO_WRITE)
             len = pwrite(aiocb->aio_fildes,
                          (const char *)buf + offset,
                          aiocb->aio_nbytes - offset,
                          aiocb->aio_offset + offset);
         else
             len = pread(aiocb->aio_fildes,
                         buf + offset,
                         aiocb->aio_nbytes - offset,
                         aiocb->aio_offset + offset);

         if (len == -1 && errno == EINTR)
             continue;
         else if (len == -1) {
             offset = -errno;
             break;
         } else if (len == 0)
             break;

         offset += len;
    }

    return offset;
}

static size_t handle_aiocb_rw(struct qemu_paiocb *aiocb)
{
    size_t nbytes;
    char *buf;

    if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
        /*
         * If there is just a single buffer, and it is properly aligned
         * we can just use plain pread/pwrite without any problems.
         */

        if (aiocb->aio_niov == 1)
             return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);

        /*
         * We have more than one iovec, and all are properly aligned.
         *
         * Try preadv/pwritev first and fall back to linearizing the
         * buffer if it's not supported.
         */

        if (preadv_present) {
            nbytes = handle_aiocb_rw_vector(aiocb);
            if (nbytes == aiocb->aio_nbytes)
                return nbytes;
            if (nbytes < 0 && nbytes != -ENOSYS)
                return nbytes;
            preadv_present = 0;
        }

        /*
         * XXX(hch): short read/write.  no easy way to handle the reminder
         * using these interfaces.  For now retry using plain
         * pread/pwrite?
         */

    }

    /*
     * Ok, we have to do it the hard way, copy all segments into
     * a single aligned buffer.
     */

    buf = qemu_memalign(512, aiocb->aio_nbytes);
    if (aiocb->aio_type & QEMU_AIO_WRITE) {
        char *p = buf;
        int i;

        for (i = 0; i < aiocb->aio_niov; ++i) {
            memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
            p += aiocb->aio_iov[i].iov_len;
        }
    }

    nbytes = handle_aiocb_rw_linear(aiocb, buf);
    if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
        char *p = buf;
        size_t count = aiocb->aio_nbytes, copy;
        int i;

        for (i = 0; i < aiocb->aio_niov && count; ++i) {
            copy = count;
            if (copy > aiocb->aio_iov[i].iov_len)
                copy = aiocb->aio_iov[i].iov_len;
            memcpy(aiocb->aio_iov[i].iov_base, p, copy);
            p     += copy;
            count -= copy;
        }
    }
    qemu_vfree(buf);

    return nbytes;
}

static void *aio_thread(void *unused)
{
    pid_t pid;

    pid = getpid();

    while (1) {
        struct qemu_paiocb *aiocb;
        size_t ret = 0;
        qemu_timeval tv;
        struct timespec ts;

        qemu_gettimeofday(&tv);
        ts.tv_sec = tv.tv_sec + 10;
        ts.tv_nsec = 0;

        mutex_lock(&lock);

        while (QTAILQ_EMPTY(&request_list) &&
               !(ret == ETIMEDOUT)) {
            ret = cond_timedwait(&cond, &lock, &ts);
        }

        if (QTAILQ_EMPTY(&request_list))
            break;

        aiocb = QTAILQ_FIRST(&request_list);
        QTAILQ_REMOVE(&request_list, aiocb, node);
        aiocb->active = 1;
        idle_threads--;
        mutex_unlock(&lock);

        switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
        case QEMU_AIO_READ:
        case QEMU_AIO_WRITE:
                ret = handle_aiocb_rw(aiocb);
                break;
        case QEMU_AIO_FLUSH:
                ret = handle_aiocb_flush(aiocb);
                break;
        case QEMU_AIO_IOCTL:
                ret = handle_aiocb_ioctl(aiocb);
                break;
        default:
                fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
                ret = -EINVAL;
                break;
        }

        mutex_lock(&lock);
        aiocb->ret = ret;
        idle_threads++;
        mutex_unlock(&lock);

        if (kill(pid, aiocb->ev_signo)) die("kill failed");
    }

    idle_threads--;
    cur_threads--;
    mutex_unlock(&lock);

    return NULL;
}

static void spawn_thread(void)
{
    sigset_t set, oldset;

    cur_threads++;
    idle_threads++;

    /* block all signals */
    if (sigfillset(&set)) die("sigfillset");
    if (sigprocmask(SIG_SETMASK, &set, &oldset)) die("sigprocmask");

    thread_create(&thread_id, &attr, aio_thread, NULL);

    if (sigprocmask(SIG_SETMASK, &oldset, NULL)) die("sigprocmask restore");
}

static void qemu_paio_submit(struct qemu_paiocb *aiocb)
{
    aiocb->ret = -EINPROGRESS;
    aiocb->active = 0;
    mutex_lock(&lock);
    if (idle_threads == 0 && cur_threads < max_threads)
        spawn_thread();
    QTAILQ_INSERT_TAIL(&request_list, aiocb, node);
    mutex_unlock(&lock);
    cond_signal(&cond);
}

static ssize_t qemu_paio_return(struct qemu_paiocb *aiocb)
{
    ssize_t ret;

    mutex_lock(&lock);
    ret = aiocb->ret;
    mutex_unlock(&lock);

    return ret;
}

static int qemu_paio_error(struct qemu_paiocb *aiocb)
{
    ssize_t ret = qemu_paio_return(aiocb);

    if (ret < 0)
        ret = -ret;
    else
        ret = 0;

    return ret;
}

static void posix_aio_read(void *opaque)
{
    PosixAioState *s = opaque;
    struct qemu_paiocb *acb, **pacb;
    int ret;
    ssize_t len;

    /* read all bytes from signal pipe */
    for (;;) {
        char bytes[16];

        len = read(s->rfd, bytes, sizeof(bytes));
        if (len == -1 && errno == EINTR)
            continue; /* try again */
        if (len == sizeof(bytes))
            continue; /* more to read */
        break;
    }

    for(;;) {
        pacb = &s->first_aio;
        for(;;) {
            acb = *pacb;
            if (!acb)
                goto the_end;
            ret = qemu_paio_error(acb);
            if (ret == ECANCELED) {
                /* remove the request */
                *pacb = acb->next;
                qemu_aio_release(acb);
            } else if (ret != EINPROGRESS) {
                /* end of aio */
                if (ret == 0) {
                    ret = qemu_paio_return(acb);
                    if (ret == acb->aio_nbytes)
                        ret = 0;
                    else
                        ret = -EINVAL;
                } else {
                    ret = -ret;
                }
                /* remove the request */
                *pacb = acb->next;
                /* call the callback */
                acb->common.cb(acb->common.opaque, ret);
                qemu_aio_release(acb);
                break;
            } else {
                pacb = &acb->next;
            }
        }
    }
 the_end: ;
}

static int posix_aio_flush(void *opaque)
{
    PosixAioState *s = opaque;
    return !!s->first_aio;
}

static PosixAioState *posix_aio_state;

static void aio_signal_handler(int signum)
{
    if (posix_aio_state) {
        char byte = 0;

        write(posix_aio_state->wfd, &byte, sizeof(byte));
    }

    qemu_service_io();
}

static void paio_remove(struct qemu_paiocb *acb)
{
    struct qemu_paiocb **pacb;

    /* remove the callback from the queue */
    pacb = &posix_aio_state->first_aio;
    for(;;) {
        if (*pacb == NULL) {
            fprintf(stderr, "paio_remove: aio request not found!\n");
            break;
        } else if (*pacb == acb) {
            *pacb = acb->next;
            qemu_aio_release(acb);
            break;
        }
        pacb = &(*pacb)->next;
    }
}

static void paio_cancel(BlockDriverAIOCB *blockacb)
{
    struct qemu_paiocb *acb = (struct qemu_paiocb *)blockacb;
    int active = 0;

    mutex_lock(&lock);
    if (!acb->active) {
        QTAILQ_REMOVE(&request_list, acb, node);
        acb->ret = -ECANCELED;
    } else if (acb->ret == -EINPROGRESS) {
        active = 1;
    }
    mutex_unlock(&lock);

    if (active) {
        /* fail safe: if the aio could not be canceled, we wait for
           it */

        while (qemu_paio_error(acb) == EINPROGRESS)
            ;
    }

    paio_remove(acb);
}

static AIOPool raw_aio_pool = {
    .aiocb_size         = sizeof(struct qemu_paiocb),
    .cancel             = paio_cancel,
};

BlockDriverAIOCB *paio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
        BlockDriverCompletionFunc *cb, void *opaque, int type)
{
    struct qemu_paiocb *acb;

    acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
    if (!acb)
        return NULL;
    acb->aio_type = type;
    acb->aio_fildes = fd;
    acb->ev_signo = SIGUSR2;
    if (qiov) {
        acb->aio_iov = qiov->iov;
        acb->aio_niov = qiov->niov;
    }
    acb->aio_nbytes = nb_sectors * 512;
    acb->aio_offset = sector_num * 512;

    acb->next = posix_aio_state->first_aio;
    posix_aio_state->first_aio = acb;

    qemu_paio_submit(acb);
    return &acb->common;
}

BlockDriverAIOCB *paio_ioctl(BlockDriverState *bs, int fd,
        unsigned long int req, void *buf,
        BlockDriverCompletionFunc *cb, void *opaque)
{
    struct qemu_paiocb *acb;

    acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
    if (!acb)
        return NULL;
    acb->aio_type = QEMU_AIO_IOCTL;
    acb->aio_fildes = fd;
    acb->ev_signo = SIGUSR2;
    acb->aio_offset = 0;
    acb->aio_ioctl_buf = buf;
    acb->aio_ioctl_cmd = req;

    acb->next = posix_aio_state->first_aio;
    posix_aio_state->first_aio = acb;

    qemu_paio_submit(acb);
    return &acb->common;
}

void *paio_init(void)
{
    struct sigaction act;
    PosixAioState *s;
    int fds[2];
    int ret;

    if (posix_aio_state)
        return posix_aio_state;

    s = qemu_malloc(sizeof(PosixAioState));

    sigfillset(&act.sa_mask);
    act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
    act.sa_handler = aio_signal_handler;
    sigaction(SIGUSR2, &act, NULL);

    s->first_aio = NULL;
    if (pipe(fds) == -1) {
        fprintf(stderr, "failed to create pipe\n");
        return NULL;
    }

    s->rfd = fds[0];
    s->wfd = fds[1];

    fcntl(s->rfd, F_SETFL, O_NONBLOCK);
    fcntl(s->wfd, F_SETFL, O_NONBLOCK);

    qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);

    ret = pthread_attr_init(&attr);
    if (ret)
        die2(ret, "pthread_attr_init");

    ret = pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
    if (ret)
        die2(ret, "pthread_attr_setdetachstate");

    QTAILQ_INIT(&request_list);

    posix_aio_state = s;

    return posix_aio_state;
}